Agentic commerce is here: will an AI agent recommend your brand, or a rival?

For twenty years the job was to win a click. A buyer searched, scanned a page of links, and chose one. Agentic commerce quietly changes the job. Now a buyer can tell an assistant "find me the best noise-cancelling headphones under 300 dollars with 30 hours of battery," and an AI agent does the searching, comparing and shortlisting, and in a growing number of flows completes the purchase, with no page of links and often no visit to anyone's website.

This is not a forecast. ChatGPT shipped in-conversation instant checkout in late 2025 through the Agentic Commerce Protocol it built with Stripe, reportedly serving on the order of hundreds of millions of weekly users, and Google announced its own Universal Commerce Protocol in early 2026 for AI Mode and Gemini. By multiple industry estimates a meaningful and fast-growing share of online transactions will involve an AI agent within a few years. Treat those size figures as directional, but the direction is not in doubt.

The question that lands on a CMO or acquisition lead's desk is simple and uncomfortable: when an agent shops your category, does it recommend you, or a competitor?

The agent's shortlist is a ranked read of who the model names for a category, and how often. This is a visibility view of category buying prompts. Illustrative data.

Getting recommended takes two layers, not one

Most of the good writing on agentic commerce is about the plumbing, and it is right. Agents do not read your blog post to decide what to recommend. They evaluate structured product data: clean catalogs, enriched metadata, JSON-LD, an agent-readable sitemap, product APIs, and support for the checkout protocols. If an agent cannot reliably read your SKU, it cannot recommend it. That work is foundational and it is a real growth lever. Get it done.

But there is a second layer that the feed conversation tends to skip, and it is the one that decides ties. Before an agent ever executes against your product API, the model has to consider you. When a shopper asks an open question rather than naming a product, the agent starts from the same place a person asking the same question gets: the set of brands the model surfaces as good answers for that category. That shortlist is a visibility question, not a feed question.

An agent can only buy from the shortlist the model builds. Complete product data gets you picked once you are on the list. Visibility is how you get on it.

So the two layers are complementary, and both are necessary. The plumbing makes you buyable. The visibility makes you considered. A brand with a flawless feed that no engine names for the category is a store with immaculate shelves on a street no one walks down.

Why the shortlist is the new scoreboard

In classic search a weak position still left you on the page. A determined buyer could scroll to result eight and click. In an agent flow there is no page to scroll. The agent synthesizes a shortlist of a few names and often acts on the top of it, with no click and no visit for you to even see in analytics. Being the fourth-best-regarded brand in the model's read of your category is no longer a soft loss of some traffic. It can be a hard absence from the only list that gets acted on.

That is why the brands treating this seriously are moving their scoreboard from rank to recommendation: not "where do I rank for this keyword," but "for the buying questions in my category, does the model name me, how often, and against whom." It is the same shift GEO has been describing for a while, with the stakes raised, because now a machine acts on the answer instead of a human weighing it.

The honest limit: you cannot see inside a personalized session

Here is where a lot of "agent visibility" claims overreach, so we will be straight about it. A specific person's agent session is logged in and personalized: it can be shaped by their history, their location, their prior purchases, and context you will never observe. Nobody credible can show you exactly what one shopper's agent recommended to them, and any tool that claims to is selling certainty it does not have.

What you can measure, honestly and repeatably, is the category-level signal: query the engines with the buying-intent prompts a shopper in your category would ask, without a logged-in personal history, and track whether they name your brand, how often, against which rivals, and which sources they lean on. Read as a rate over a rolling window with a confidence interval, that is a real, defensible measure of whether the model considers you a good answer. It is the discovery layer that feeds the shortlist, and it is the part you can actually move. Treat it as what it is, a strong directional read of your standing, not a claim to have watched a stranger's checkout.

What to measure now

• Your category buying prompts. The open, unbranded questions a shopper asks before they know which brand they want. These are the prompts an agent starts from.
• Whether the model names you, and how often. Not once, but as a rate over a window, because answers are non-deterministic.
• Who it names instead. Your share of voice against the specific rivals that keep showing up on the shortlist.
• Which sources it cites. The reviews, roundups and pages the engine leans on to build the shortlist, so you know where to earn presence.
• Movement. Whether all of the above is trending up or down as you ship changes, refreshed often enough to connect cause to effect.

How llemmy helps

llemmy measures the visibility layer of agentic commerce, precisely and honestly. It runs your category and buying-intent prompts against ChatGPT, Claude, Gemini, Perplexity and Google AI, and tracks whether each engine names your brand, how often, and against which competitors, as a rate over a rolling window with a 95% confidence interval and a sample size on every number. It shows your share of voice on the shortlist and the sources each engine cites to build it, refreshed daily so a change in who the model recommends shows up quickly. What it does not do, by design, is pretend to see inside a personalized logged-in agent session. It reads the official model APIs and Google's AI surfaces at the category level, which is the honest, movable signal for whether the model considers you before an agent ever reaches your feed. Pair that with the product-data and protocol work, and you can see both halves of whether an agent will recommend you. Run a free GEO audit or start tracking free to see whether AI engines name your brand for your category.

FAQ

What is agentic commerce?

It is when an AI agent handles the shopping on a person's behalf: it takes a request like "best noise-cancelling headphones under 300 dollars," searches product data and reviews, builds a shortlist, and in some flows completes the purchase without the buyer visiting a website. ChatGPT and Google have both shipped agentic checkout flows, so the shortlist the model builds increasingly decides the sale.

How do brands get recommended by AI shopping agents?

Two layers have to work. The plumbing layer is machine-readable product data: structured data, clean catalogs, and support for the agent checkout protocols. The visibility layer is whether AI answers already name your brand as a good answer for the category and buying-intent queries a shopper asks. The agent picks from the shortlist the model surfaces, so if the model does not name you, complete product data alone will not put you in the cart.

Can you measure whether AI agents recommend your brand?

You can measure the visibility layer directly and honestly: query the AI engines with the category and buying-intent prompts a shopper would use, and track whether they name your brand, how often, against which rivals, and which sources they cite, as a rate over a rolling window with a confidence interval. You cannot see inside a specific person's logged-in, personalized agent session, and any honest measurement should say so rather than pretend to.

Is agentic commerce visibility different from SEO?

It overlaps but the unit changes. SEO optimizes for a ranked list of links a human scans. Agentic commerce optimizes for being named in a synthesized shortlist an agent acts on, often with no click and no website visit. You still need the technical foundations, but the new scoreboard is whether the model recommends you, which is what AI visibility measurement tracks.

By the llemmy team, July 2026. Grounded in public reporting on ChatGPT Instant Checkout / the Agentic Commerce Protocol (OpenAI and Stripe) and Google's Universal Commerce Protocol, plus industry estimates of agentic commerce growth, which vary by source and should be read as directional. Related reading: How to track your brand across AI engines, Share of voice in AI answers, and The AI Citation Gap.